FPRD Framework-Shape Probe

Arithmetic dynamics, local witnesses, and the possible shape of an FPRD reasoning machine

Date: 2026-07-27
FPRD reference point: repository commit b70ff6ccb4d333c825c598bf52f054db72d24cbb

Purpose

FPRD is still determining its own mathematical shape. The relevant question is therefore not only whether Collatz, Goldbach, or reverse-and-add should be “promoted” as research lanes. These problems can instead be used to test:

• what an FPRD object must contain;
• which finiteness and compactness assumptions are substantive;
• what finite observations can and cannot establish;
• what it means to admit a finite presentation without making the concept vacuous;
• and what an eventual FPRD reasoning machine could produce.

This report develops two exact examples and one structural stress test. It then proposes specific revisions to the framework and a bounded design for a reasoning machine.

No progress on the Collatz or Goldbach conjectures is claimed. The exact results here concern their FPRD formulations.

Executive assessment

Yes, an FPRD-based machine could plausibly say something mathematically interesting about Collatz-like and reverse-and-add systems. The likely outputs are initially:

• exact finite-shadow formulas;
• certified families of residue classes with common orbit behavior;
• automatically discovered semiconjugacies or orbit schemas;
• resource-escape laws;
• no-lift obstructions explaining why local evidence cannot prove the global statement;
• and compressed induction certificates over infinite input families.

That is a meaningful theorem-discovery capability. It is not yet a general solver for famous conjectures.

The arithmetic probes reveal four missing or underemphasized parts of FPRD:

1. A distinguished admissible locus.
   Arithmetic truth often lives on $\mathbb N$ or $\mathbb Z$, while the useful finite-observation completion contains additional $p$-adic or profinite points. FPRD must record both.

2. Property-relative observation adequacy.
   Observations can separate exact points yet remain useless for a target predicate such as eventual reachability or primality.

3. Parametric witness semantics.
   Goldbach has the form $\forall n\,\exists(p,q)$, not merely membership of one fixed behavior. Witness fibers and uniformity in the parameter need to be visible.

4. A nonvacuous presentation doctrine.
   Every computable arithmetic map has a finite program. If that fact alone counts as an FPRD finite presentation, the designation has almost no mathematical content. The doctrine, primitives, normalization, resource strata, and shadow compiler must be declared.

The best immediate change is not a broad axiomatization. Add an optional admissible-locus layer to the current compact observation tower, together with a short adequacy ledger. Test it on the Collatz example below.

1. Where the current FPRD shape is strong

The current repository already makes several decisions that should be kept.

Presentation is doctrine-relative

M33 does not treat “finite presentation” as an absolute property. It fixes a presentation doctrine $\mathsf P$, a semantic map

$$\llbracket-\rrbracket:\mathsf P\to X,$$

and a target-independent resource map

$$\kappa:\mathsf P\to R.$$

Semantic resource strata, rather than raw syntax size, prevent state renaming, unreachable padding, and gauge changes from masquerading as mathematical resource growth.

This is the correct basis for arithmetic work. It needs to become an enforced admission requirement across FPRD, not just an M33-specific safeguard.

Multiple resources are already allowed

M33’s resource field is partially ordered and can contain incomparable coordinates. Probe Set 2 independently found that transduction and probabilistic recognition may require input depth, output resolution, and numerical precision as separate resources.

Arithmetic adds further coordinates:

• iteration depth;
• integer height or digit length;
• modulus or residue resolution;
• witness height;
• and sometimes proof-certificate size.

The resource field is therefore a strong part of the current architecture.

Sewing and escape are the right central questions

FPRD-T60 and T61 distinguish:

$$\text{one bounded realization at every finite horizon}$$

from

$$\text{one bounded realization of the exact object}.$$

The compact-stratum theorem explains when the two coincide. The escape theorem explains how horizon-fitting realizations leave every compact core when they do not.

Arithmetic examples fit this question unusually well, but they also reveal that the exact arithmetic domain may be only a dense, noncompact subset of the completion.

2. The principal mismatch: compact completion versus arithmetic locus

The M22 finite-observation tower begins with a compact Hausdorff exact carrier $X$, finite quotients $X_N$, and separating observation maps

$$q_N:X\to X_N.$$

This is mathematically clean. It makes exact reconstruction from a compatible thread possible.

For arithmetic dynamics, however, the natural situation is often:

$$A\hookrightarrow\widehat X,$$

where:

• $A$ is the intended arithmetic state set, such as $\mathbb N_{>0}$;
• $\widehat X$ is a compact observation completion, such as $\mathbb Z_2$;
• the image of $A$ may be dense but not closed in $\widehat X$;
• the action on $A$ extends continuously to $\widehat X$;
• and the target statement quantifies only over $A$.

Compatible finite observations reconstruct a point of $\widehat X$. They need not reconstruct a point of $A$.

This is not a minor technical issue. For Collatz, it is exactly where a complete symbolic description of the 2-adic dynamics stops short of resolving the positive-integer conjecture.

Proposed minimal extension

Replace the implicit identification of “exact carrier” and “completed carrier” by the optional data

$$(\widehat X,A,\iota),$$

where:

• $\widehat X$ is the compact finite-observation completion;
• $A$ is the admissible locus;
• $\iota:A\hookrightarrow\widehat X$ is an injective semantic map;
• and each declared action specifies whether it preserves $A$, extends to $\widehat X$, or both.

The original M22 setting is recovered by taking $A=\widehat X$.

This one addition allows FPRD to state separately:

1. a finite thread has a unique completion point;
2. that point is or is not arithmetically admissible;
3. bounded admissible approximants do or do not exist;
4. any approximants must or must not escape in height.

3. Exact Collatz shape probe

3.1 The accelerated map and parity observations

Use the standard accelerated Collatz map

$$T(n)= \begin{cases} n/2,&n\equiv0\pmod2,\\[3pt] (3n+1)/2,&n\equiv1\pmod2. \end{cases}$$

For a starting value $x$, define its parity sequence

$$Q(x)= \bigl( x\bmod2,\, T(x)\bmod2,\, T^2(x)\bmod2,\ldots \bigr).$$

The map extends to the 2-adic integers. The parity sequence is a bijective coding of $\mathbb Z_2$ by infinite binary sequences, and under this coding the Collatz action becomes the left shift. This classical 2-adic conjugacy is reviewed explicitly by Rozier, “Parity Sequences of the $3x+1$ Map on the 2-adic Integers and Euclidean Embedding”, and originates in the Bernstein–Lagarias conjugacy work.

At finite horizon $N$, two integers have the same first $N$ parity bits if and only if they are congruent modulo $2^N$. Consequently:

$$\mathbb Z/2^N\mathbb Z \;\cong\; \{0,1\}^N.$$

Every finite binary parity word is realized by infinitely many positive integers, while every infinite parity sequence is realized by exactly one 2-adic integer.

3.2 FPRD formulation

• Admissible locus: $A=\mathbb N_{>0}$.
• Compact completion: $\widehat X=\mathbb Z_2$, or equivalently $\{0,1\}^{\mathbb N}$ under parity coding.
• Action: $T$, conjugate on the completion to the shift.
• Finite observation: the first $N$ parity bits.
• Arithmetic height: $$h(n)=\lceil\log_2(n+1)\rceil.$$
• Height-$r$ core: $$A_{\le r} =\{n\in\mathbb N_{>0}:h(n)\le r\}.$$

Every height core is finite. The entire arithmetic locus is noncompact and dense in the completion.

3.3 Derived result: an exact natural escape profile

Consider the completed point whose parity sequence is

$$s=1^\omega.$$

This point is the 2-adic integer $-1$, since

$$T(-1)=\frac{3(-1)+1}{2}=-1.$$

It is not a positive integer. Nevertheless, its first $N$ parity bits are realized by positive integers.

The first $N$ bits are all $1$ exactly when

$$n\equiv-1\pmod{2^N}.$$

The smallest positive representative is

$$n_N=2^N-1.$$

It can also be checked directly. For $0\le k\le N$,

$$T^k(2^N-1) =3^k2^{N-k}-1.$$

For $k<N$, this is odd, so the first $N$ parity observations are all $1$.

The height is exact:

$$h(n_N) =\left\lceil\log_2(2^N)\right\rceil =N.$$

Therefore the least admissible height required to fit the target through horizon $N$ is

$$\boxed{d_s(N)=N.}$$

No positive integer fits the exact infinite target:

$$\boxed{d_s(\infty)=\infty.}$$

For the height-$r$ core, the longest all-$1$ prefix has length $r$, so the first observation horizon excluding that core is

$$\boxed{e_s(r)=r+1.}$$

The calculation was exhaustively checked through $N=12$, although the formulas above are direct proofs for every $N$.

3.4 What this tells FPRD

This is a natural arithmetic instance of compact-exhaustion escape:

• every finite horizon is matched inside the arithmetic locus;
• no exact arithmetic point matches;
• the approximating positive integers leave every fixed height core;
• and the resource demand and escape modulus are both exact.

It also shows why the admissible locus cannot be omitted. In the completion, the observations sew perfectly to $-1$. For the positive-integer problem, that sewn point is inadmissible.

The distinction should be named:

$$\textbf{completion realization} \quad\ne\quad \textbf{admissible realization}.$$

3.5 Relation to the actual Collatz conjecture

The usual conjecture says that every positive integer eventually reaches the cycle $1\leftrightarrow2$. In parity coding, the cycle has tail

$$(10)^\omega.$$

Thus the conjecture can be written as

$$Q(\mathbb N_{>0}) \subseteq \{u(10)^\omega:u\in\{0,1\}^*\}.$$

The completed system contains every binary sequence. Most completion points do not have an eventually alternating parity tail. The difficult information is therefore not the completed shift dynamic; that dynamic is completely understood. It is the arithmetic shape of the dense subset

$$Q(\mathbb N_{>0}) \subset\{0,1\}^{\mathbb N}.$$

This gives a precise warning:

A perfect FPRD description of the completed observation dynamic need not resolve a theorem about the admissible arithmetic locus.

For Collatz, an FPRD machine must learn or exploit structure beyond parity prefixes.

3.6 A concrete machine-level Collatz output

Finite parity cylinders do support exact descent certificates.

For a parity word

$$\varepsilon=(\varepsilon_0,\ldots,\varepsilon_{N-1}), \qquad s=\sum_i\varepsilon_i,$$

composition of the affine Collatz branches gives

$$T^N(n) = \frac{3^s n+c_\varepsilon}{2^N},$$

where

$$c_\varepsilon = \sum_{j=0}^{N-1} \varepsilon_j\,2^j\, 3^{\sum_{i=j+1}^{N-1}\varepsilon_i}.$$

Each parity word identifies one residue class modulo $2^N$. The machine can therefore produce a certificate of the form:

1. this residue class has parity word $\varepsilon$;
2. this is its exact affine $N$-step iterate;
3. $3^s<2^N$;
4. hence $T^N(n)<n$ above the explicit threshold $$n>\frac{c_\varepsilon}{2^N-3^s}.$$

At horizon $2$, the four cylinders are:

An FPRD refinement engine could retain only unresolved cylinders, refine them to longer parity words, synthesize affine descent certificates, and search for a compressed rule covering an infinite family.

This is close to established stopping-time and residue-class methods, so it would require a strict comparator audit. The possible FPRD contribution would be:

• an exact demand profile for the unresolved cylinders;
• a compressed obligation algebra for refinement;
• a transferable certificate format;
• or a new uniform family, not merely a larger verified bound.

This is a credible route by which an FPRD machine could say something interesting about Collatz without pretending that finite checking proves the conjecture.

4. Exact Goldbach shape probe

4.1 Goldbach is a witness problem

For an even integer $n\ge4$, define the witness fiber

$$W(n) = \{(p,q):p,q\text{ prime and }p+q=n\}.$$

The conjecture is

$$\forall n\in2\mathbb N,\ n\ge4: \quad W(n)\ne\varnothing.$$

Encoding only the Boolean language of successful $n$ values discards much of the structure. A useful FPRD formulation should retain:

• the parameter $n$;
• the witness fiber $W(n)$;
• observation maps on candidate witnesses;
• witness height;
• and uniformity of the construction as $n$ grows.

4.2 A local modular abstraction

At modulus $m$, a necessary local approximation to primality is being a unit modulo $m$. Define the modular witness shadow

$$U_m(n) = \{(a,b)\in(\mathbb Z/m\mathbb Z)^{\times 2}: a+b\equiv n\pmod m\}.$$

4.3 Derived result: every even integer has unit witnesses at every modulus

For every even $n$ and every $m\ge2$,

$$\boxed{U_m(n)\ne\varnothing.}$$

Proof

Factor $m$ into prime powers and work componentwise.

• For $2^e$, choose $a=1$. Since $n$ is even, $b=n-1$ is odd. Both are units modulo $2^e$.
• For an odd prime power $p^e$, choose $a\bmod p$ different from both $0$ and $n\bmod p$. This is possible because $p\ge3$. Then neither $a$ nor $n-a$ is divisible by $p$, so both lift as units modulo $p^e$.
• Apply the Chinese remainder theorem across the prime-power components.

The statement was also checked computationally for every $2\le m\le200$ and even $4\le n\le100$.

4.4 Why sewing the local witnesses is insufficient

The modular unit solutions can be chosen compatibly and sewn into profinite units

$$\alpha,\beta\in\widehat{\mathbb Z}^{\times}$$

with

$$\alpha+\beta=n.$$

This is a valid global solution in the profinite completion. It does not produce ordinary prime integers.

The local tower has answered a different question:

$$\text{Are there local unit obstructions?}$$

For even $n$, the answer is no. Goldbach asks:

$$\text{Are there bounded integral witnesses satisfying primality?}$$

Point separation of the profinite observation tower does not repair the mismatch. The predicate “is prime” has been replaced by the weaker predicate “is locally a unit.”

4.5 The framework distinction

FPRD should distinguish:

1. point adequacy: observations separate completed points;
2. action adequacy: the exact action factors coherently through the observations;
3. predicate adequacy: the observations preserve enough information to decide the declared readout;
4. witness adequacy: compatible observed witnesses lift to an admissible witness;
5. uniform adequacy: the lifting or certificate works uniformly over the problem parameter.

The Goldbach unit tower has point and algebraic compatibility. It is not predicate- or witness-adequate for primality.

4.6 Height changes the fixed-instance picture

For one fixed $n$, every positive Goldbach witness automatically satisfies

$$2\le p,q\le n-2.$$

The candidate witness set is finite. Sufficiently detailed exact observations can simply inspect primality for every candidate. There is no fixed-instance compactness mystery.

The conjecture is difficult because the parameter $n$ is unbounded. The witness core and the verification work grow with $n$.

This exposes a second missing distinction:

$$\textbf{one fixed target} \quad\ne\quad \textbf{a uniformly presented target family}.$$

M33’s demand field is formulated for a fixed target $x$ as observation horizon increases. Goldbach calls for a family-level demand surface, for example

$$\mathcal F(n,b)$$

or

$$d(n,b),$$

where $n$ is the problem parameter and $b$ is the observation budget.

The machine should search for a uniform law in $n$, not regard independent finite verification for each $n$ as a sewn theorem.

4.7 What an FPRD machine might say about Goldbach

Initially, plausible outputs include:

• exact local obstruction criteria for generalized additive-prime patterns;
• compatible modular witness constructions;
• automatic derivation of local density factors;
• identification of observation systems that cannot distinguish primes from profinite units;
• resource lower bounds for any certificate family in a chosen doctrine;
• and a formal no-lift result explaining why a proposed modular proof cannot establish the conjecture.

The last item is mathematically useful. A machine that reliably explains why a tempting finite-shadow argument cannot reach Goldbach is already doing more than brute-force verification.

Actual progress on Goldbach would require an admitted doctrine carrying analytic information about primes, not only congruence data. FPRD should make that dependency explicit rather than hiding it in the word “observation.”

5. Reverse-and-add and the admission problem

Let

$$R_b(n)=n+\operatorname{rev}_b(n),$$

where $\operatorname{rev}_b$ reverses the base-$b$ digits, and let the readout test whether the current value is palindromic.

This looks immediately compatible with FPRD:

• a finitely described update rule;
• an infinite orbit;
• finite digit observations;
• a target readout;
• iteration-depth and digit-length resources.

It also exposes a possible weakness.

A finite program is too permissive by itself

The reverse-and-add rule has a finite implementation. So do Collatz, arbitrary Turing machines, and every computable map.

If “has a finite program” is sufficient for admission as a mathematically meaningful finite presentation, then almost every computable dynamical system qualifies. FPRD remains a valid meta-framework, but finite presentation no longer distinguishes a tractable or compressible class.

The correct statement is:

A presentation is finite only relative to a declared doctrine and resource semantics; finiteness alone carries no tractability conclusion.

M33 already says that its theorems are doctrine-relative. The arithmetic probe suggests turning that warning into a universal admission rule.

A useful reverse-and-add doctrine

A nonvacuous probe might declare:

• states represented by finite digit words;
• primitive operations realized by a specified class of digit transducers;
• reversal access explicitly allowed or charged;
• carries represented by finite control;
• a normal form identifying semantically equivalent transducers;
• digit length and control size as resources;
• observation budgets for iteration depth and exposed digit positions.

The first question is then not whether reverse-and-add is computable. It is:

Does the orbit or target behavior admit a bounded or regularly growing presentation in this digit doctrine?

A realistic machine target

An FPRD machine could:

1. compute a long reverse-and-add orbit;
2. infer a regular expression or parameterized digit-word schema;
3. compile one update step into a finite carry transducer;
4. prove that the schema is invariant under the update;
5. prove that no word in the schema is a palindrome.

That would yield an exact infinite-family theorem, not merely a longer computation. It is the appropriate shape for studying Lychrel behavior in bases where regular orbit schemas exist.

For base $10$ and the $196$ sequence, this probe did not identify an established regular orbit schema; that point would need a dedicated literature audit before being stated exhaustively. The FPRD contribution would have to be the discovery or exclusion of a compressed orbit invariant. Digit growth by itself is not evidence.

6. Proposed FPRD object shape

The arithmetic examples suggest the following object, but it should first be implemented as an interface dossier rather than a new foundational axiom system.

An admissible presented observation system consists of:

$$\mathfrak X= ( \mathsf P, \widehat X, A, \iota, \mathsf Act, \{q_b\}_{b\in B}, \kappa, h, \mathsf O ).$$

The components are:

Not every application needs every component. In ordinary regular-language recognition:

• $A=\widehat X$;
• $B=\mathbb N$;
• no separate height may be needed.

For Collatz:

• $A=\mathbb N_{>0}$;
• $\widehat X=\mathbb Z_2$;
• $B$ includes parity horizon;
• $h$ is integer height.

For Goldbach:

• $A$ includes integral witnesses;
• $B$ may include modulus, magnitude, and analytic resolution;
• $\mathsf O$ is a parameterized witness relation.

Do not require compactness of the admissible locus

Compactness should remain a theorem hypothesis on the relevant stratum, not a universal property of $A$. Arithmetic loci are often noncompact. Instead, require a declared exhaustion

$$A_{\le0}\subseteq A_{\le1}\subseteq\cdots, \qquad A=\bigcup_r A_{\le r},$$

with each core compact or finite when an escape theorem needs it.

This preserves the current M33 logic.

7. Observation adequacy ledger

Each admitted probe should state which adequacy properties it has. These are not all-or-nothing grades.

A1. Point separation

$$\bigl(\forall b,\ q_b(x)=q_b(y)\bigr)\Longrightarrow x=y \quad \text{in }\widehat X.$$

This is the current M22 requirement.

A2. Action exposure

For each action, state how much additional source observation is required to compute a target observation. This generalizes the M22 exposure bound to a directed budget poset.

A3. Locus completeness

Determine whether

$$\varprojlim_b q_b(A)$$

contains only images of points of $A$, or also completion artifacts.

Equivalently, record whether $A$ is closed in the chosen completion and, if not, provide a height exhaustion and escape interpretation.

Collatz positive integers fail locus completeness in the parity completion.

A4. Readout stability

State whether the readout extends continuously or at least as a closed predicate to the completion.

Probe Set 2’s probabilistic cutpoint example fails this condition. Eventual Collatz reachability also is not a finite-cylinder property.

A5. Witness lifting

For relational problems, state whether compatible finite witness shadows lift to an admissible exact witness and under what resource bound.

Goldbach’s unit shadows fail to lift to prime witnesses.

A6. Uniform effectivity

Provide algorithms, uniform in the budget, for:

• constructing the finite shadow;
• applying the finite action;
• checking bonding;
• decoding a positive certificate;
• and producing a counterexample or escape certificate when lifting fails.

Without uniform effectivity, finite shadows may exist mathematically but not support a reasoning machine.

8. Revised admission process

Admission should not be one binary decision. A proposal should answer two independent questions:

1. What does it contribute to the shape or testing of FPRD?
2. How much additional mathematical work should be spent on it now?

Required admission packet

Every new probe should provide:

1. External object.
   The original mathematical problem, without replacing it by an easier internal analogue.

2. Doctrine.
   Allowed presentation primitives, semantic equivalence, and normalization.

3. Exact locus and completion.
   Intended states, ambient completion, and any additional completion points.

4. Action.
   Exact update laws and preservation of the locus.

5. Observation budget.
   Directed resource axes, finite shadows, bonding, and separation.

6. Resource and height.
   Target-independent presentation resources and compact/finite cores.

7. Readout or witness semantics.
   What exact proposition is being studied.

8. Adequacy ledger.
   A1–A6 above, including known failures.

9. Comparator.
   The area’s native formalism and the exact proposed FPRD delta.

10. Expected artifact.
    The theorem, invariant, transfer, obstruction, algorithm, or boundary case the probe could produce.

Placement after admission

“Stop” should mean stop spending resources. It should not erase a useful boundary example.

9. Shape of an FPRD reasoning machine

The machine should not initially be framed as a general automated mathematician. A realistic first version is a certificate-producing abstraction and invariant workbench.

It would have six components.

9.1 Exact problem adapter

The adapter declares:

• the admissible locus;
• action rules;
• readout or witness relation;
• the presentation doctrine;
• and exact decoding back to the external problem.

This prevents success on a self-created surrogate from being reported as progress on the original problem.

9.2 Shadow compiler

Given a budget $b$, construct:

• the finite observation carrier;
• the abstract action;
• bonding maps;
• exposure requirements;
• and soundness certificates.

This resembles abstract interpretation, whose foundational formulation constructs sound approximations of program semantics; see Cousot and Cousot, “Abstract Interpretation: A Unified Lattice Model for Static Analysis”.

FPRD should not claim finite abstraction itself as new.

9.3 Refinement engine

When an observation is inconclusive:

1. identify the unresolved cylinder or witness class;
2. refine only that portion;
3. track which resource coordinate increased;
4. retain a certificate explaining the refinement.

This has a clear comparator in counterexample-guided abstraction refinement; see Clarke, Grumberg, Jha, Lu, and Veith, “Counterexample-Guided Abstraction Refinement”.

The FPRD-specific opportunity is to combine refinement with exact presentation demand and escape profiles.

9.4 Invariant and schema synthesizer

Search for:

• affine or linear action formulas;
• stable residue-class partitions;
• monotone quantities;
• periodic or regular digit schemas;
• semiconjugacies;
• finite bases of obligations;
• and polynomial or exponential shadow-growth laws.

Any candidate must be emitted with an induction or algebraic proof obligation, not accepted from numerical fit.

9.5 Sewing and escape auditor

For every proposed global conclusion, ask:

• Do compatible shadows lift in the completion?
• Does the lift lie in the admissible locus?
• Does the readout extend to it?
• Are witnesses uniformly bounded?
• If not, which resource escapes and at what modulus?

This is the component most characteristic of the current FPRD program.

9.6 Proof-certificate exporter

The machine’s outputs should have a small number of exact types:

• finite quotient and factorization certificate;
• exposure bound;
• invariant or semiconjugacy;
• uniform family construction;
• resource lower bound;
• escape modulus;
• no-lift obstruction;
• or externally decoded theorem.

This makes independent verification possible.

10. What the machine could realistically produce

Collatz

Plausible:

• certified affine formulas for parity cylinders;
• compressed descent covers of residue classes;
• exact demand profiles for unresolved cylinders;
• new sufficient sets or generalized-Collatz decider families;
• a proof that a proposed finite-observation argument necessarily loses the positive-integer locus.

Not supplied by current FPRD:

• a mechanism forcing every positive orbit into the $1\leftrightarrow2$ cycle;
• a new height monotone;
• or a characterization of $Q(\mathbb N_{>0})$ strong enough to prove the conjecture.

Goldbach

Plausible:

• local obstruction classifications;
• modular witness schemas;
• local density calculations;
• exact diagnoses of failed local-to-global proofs;
• resource profiles for restricted additive-prime doctrines.

Not supplied by current FPRD:

• the analytic control of prime correlations needed for the full conjecture.

Reverse-and-add and Lychrel-type systems

Plausible:

• discovery and proof of regular digit-orbit schemas;
• exact carry-state presentations;
• infinite nonpalindromic families in selected bases;
• lower bounds showing that a chosen finite-state digit doctrine cannot represent a candidate orbit at bounded resource.

Not supplied by current FPRD:

• a reason that the base-$10$ orbit of $196$ must or must not reach a palindrome.

11. Specific changes recommended now

Change 1: add an admissible-locus dossier

Add a short optional interface to the existing observation-tower documentation:

$$A\hookrightarrow\widehat X.$$

Require action preservation, closure status, height exhaustion, and a statement of whether completion points are semantically admissible.

Use the Collatz all-$1$ example as the first test.

Change 2: generalize the observation index from $\mathbb N$ to a directed poset

Retain the existing one-dimensional tower as the basic case. Permit budgets such as

$$b=(\text{time},\text{modulus},\text{height},\text{precision}).$$

Exposure becomes a monotone budget transformer rather than always an additive integer.

Change 3: add the six-entry adequacy ledger

Do not require every probe to satisfy every property. Require it to state which ones hold. Failures can be valuable boundary results.

Change 4: make uniform families explicit

Add a lightweight parametric layer for targets $x_\lambda$ and witness fibers $W(\lambda)$. Record:

• whether the presentation is uniform in $\lambda$;
• how resources scale with $\lambda$;
• and whether a single certificate schema covers all parameters.

This is necessary for Goldbach-like universal statements.

Change 5: enforce doctrine admission everywhere

“Finite program” is not an adequate doctrine. Require:

• named primitives;
• semantic normalization;
• target-independent resources;
• compact or finite bounded strata when used;
• and a uniform shadow compiler.

Change 6: separate framework contribution from research priority

Replace a single promote/stop field by:

• FPRD contribution: core-shaping, theorem candidate, transfer candidate, calibration, boundary, or uninformative;
• current expenditure: active, one-more-probe, dormant, or stop.

Change 7: add arithmetic negative controls

Keep three permanent tests:

1. Collatz completion artifact: every finite parity word is admissible, but an infinite thread may be only 2-adic.
2. Goldbach unit artifact: all modular unit shadows have witnesses, but primality does not lift.
3. Queue artifact: finite computable shadows can still encode universal computation.

Any general sewing or machine-capability claim should state why it does not fail on these controls.

12. Bounded next work

Do not begin a general arithmetic-FPRD program. Run three small contracts.

Contract A: admissible-locus extension

Formalize the Collatz all-$1$ calculation:

$$d_s(N)=N,\qquad e_s(r)=r+1,\qquad d_s(\infty)=\infty.$$

Show exactly how it instantiates or extends D28/T61. Determine whether the existing M33 definitions already express the result by choosing the arithmetic locus as a noncompact semantic stratum, or whether an explicit ambient/locus pair is required.

Success: one minimal interface revision and one exact theorem/example.

Contract B: Collatz cylinder certificate prototype

Implement the parity-cylinder affine formula and refinement engine. Ask it to produce:

• exact residue-to-parity maps;
• descent thresholds;
• a prefix tree of unresolved classes;
• and a compressed certificate for one established stopping-time coverage result.

Success: reproduce a nontrivial known result in the field’s own terms and identify one FPRD quantity not already present in the comparator.

Contract C: Goldbach witness-adequacy test

Formalize the unit-shadow lemma, the profinite witness, and the failure to lift to prime witnesses. Define the smallest useful witness-adequacy interface.

Then test one generalized additive problem where modular obstruction is actually decisive, so the same machinery has both a positive and a negative control.

Success: a reusable no-lift certificate or witness-sewing theorem.

Reverse-and-add should follow only after the digit-transducer doctrine is specified tightly enough that finite presentation is not synonymous with a finite program.

Bottom line

FPRD could become a useful mathematical reasoning system, but its likely strength is now clearer.

It is well positioned to reason about the boundary between:

$$\text{finite observable structure}$$

and

$$\text{exact global or arithmetic realization}.$$

Collatz, Goldbach, and reverse-and-add all live near that boundary, but in different ways:

• Collatz separates a completely understood symbolic completion from a poorly understood arithmetic locus.
• Goldbach separates compatible local witnesses from admissible global witnesses.
• Reverse-and-add separates finite computability from bounded structural presentation.

These examples suggest that FPRD should not be narrowed to regular-language recognition. They also suggest that it should not expand by declaring every finite program an equally informative presentation.

The right next shape is:

a doctrine-relative framework for exact dynamics on an admissible locus, finite observations in an ambient completion, typed resource demand, and certified analysis of when local behavior lifts, escapes, or fails to preserve the target property.

That shape could support a real reasoning machine. The Collatz all-$1$ escape law and the Goldbach unit-witness obstruction are suitable first arithmetic calibrations because each produces an exact result while also showing what the framework must not assume.